Módulo 6: Industry Standards and Frameworks

NIST AI RMF: Map Function

Siguiente →

Descripción

Map function te ayuda a entender el contexto de tu sistema AI: quiénes son los usuarios, qué decisiones toma el sistema, qué riesgos pueden ocurrir. Es la fase de descubrimiento e identificación antes de medir o gestionar.

Sin Map, gastás tiempo midiendo cosas que no importan y missing cosas críticas. Map te enfoca.

Al terminar vas a poder:

  • Categorizar el contexto del sistema AI usando NIST framework
  • Identificar stakeholders, impactos y riesgos
  • Clasificar AI risks por categoría (technical, societal, legal)
  • Producir un context document para tu sistema

Los 5 categorías de Map

Map 1: Context Established

Documentás:

  • Purpose: ¿qué problema resuelve el AI?
  • Users: quién interactúa (directly y indirectly)
  • Deployment context: production environment, geographies, scale
  • Lifecycle stage: planning, development, evaluation, deployment, operation
  • Use cases: específicos, no genéricos

Map 2: Categorization

Clasificás el AI system:

  • AI lifecycle phase
  • Application sector (healthcare, finance, education, etc.)
  • Specific tasks (classification, generation, recommendation)
  • Decision type (binary, multi-class, continuous, generative)

Map 3: AI Capabilities and Risks

Identificás:

  • Capabilities del sistema (qué hace)
  • Limitations conocidas
  • Risks que pueden materializar

Map 4: Impacts

Para cada stakeholder, ¿cuáles son los impactos potenciales?

  • Direct users: decisiones que afectan a quienes interactúan
  • Indirect affected: otros que no usan pero se impactan
  • Societal: efectos broader (employment, bias amplification)

Map 5: Risk identification

Lista comprehensive de riesgos posibles:

  • Performance risks: model fails to perform
  • Robustness risks: fails under unexpected inputs
  • Security risks: adversarial attacks, data exfiltration
  • Privacy risks: data exposure
  • Bias risks: unfair outcomes
  • Explainability risks: cannot justify decisions

Risk categories específicas para AI

1. Technical Risks
   - Hallucination (LLMs make things up)
   - Distribution shift (model performs differently on new data)
   - Adversarial inputs (intentional manipulation)
   - Brittleness (small input changes → big output changes)
   - Catastrophic forgetting (in continually updated models)

2. Operational Risks
   - Insufficient monitoring → unnoticed degradation
   - Cost explosion (LLM tokens, GPU resources)
   - Latency exceeding SLA
   - Dependency on third-party providers

3. Privacy Risks
   - Training data leakage
   - Inference of sensitive attributes
   - Profile inference enabling identification
   - Cross-tenant data exposure

4. Fairness Risks (covered M2)
   - Demographic disparities
   - Disparate impact on protected groups
   - Feedback loops amplifying bias

5. Transparency Risks
   - Lack of explainability for affected individuals
   - Lack of disclosure (users don't know they're talking to AI)
   - Lack of human oversight options

6. Legal/Compliance Risks
   - GDPR violations (Art. 22, consent, transfers)
   - EU AI Act non-compliance
   - Discrimination law violations
   - IP infringement (training on copyrighted material)

7. Societal Risks
   - Worker displacement
   - Concentration of decision-making power
   - Amplification of misinformation
   - Environmental impact (compute)

Aplicación: Knowledge Assistant Map document

# Map Document — AI Knowledge Assistant

## Context (Map 1)

**Purpose**: Allow employees of client organizations to quickly find
information from internal knowledge bases through Slack/Discord.

**Users**:
- Direct: employees (≈10K-100K across 50 client orgs Year 1)
- Indirect: company customers (whose data may be in KB)
- Stakeholders: org admins, security teams, executives

**Deployment**: Production B2B SaaS. Multi-region (US, EU). Multi-tenant.

**Lifecycle**: Operation (deployed, monitoring continuously)

**Use cases**:
- Tech docs lookup
- Procedural questions
- Decision support
- NOT: hiring, firing, financial approvals, medical decisions

## Categorization (Map 2)

- **Phase**: Operation
- **Sector**: Cross-industry B2B SaaS
- **Tasks**: Question-answering with retrieval; generative responses
- **Decision type**: Generative (text); occasional categorization (intent)

## Capabilities & Risks (Map 3)

**Capabilities**:
- Answer factual questions from internal docs
- Cite sources
- Handle multi-turn conversations
- Multi-language (initially Spanish + English)

**Limitations**:
- Cannot answer questions not in knowledge base
- May hallucinate if context insufficient
- Confidence calibration imperfect
- Real-time data (very fresh updates) may not be reflected

## Impacts (Map 4)

**Direct users**:
- Saved time (positive)
- Productivity (positive)
- Risk: bad answers leading to wrong actions

**Indirect affected**:
- Customers of clients (if KB contains customer-related info)
- Employees not getting attention if AI does most help

**Societal**:
- Potential displacement of help desk roles
- Productivity gain at scale

## Risk Identification (Map 5)

Top 15 risks identified:

1. Hallucination — fabricated information
2. Stale information — KB updates not reflected
3. Tenant data leakage — Client A sees Client B data
4. Training data leakage — model memorizes confidential
5. Bias in responses — different quality for different groups
6. Insufficient explainability — user can't verify answer
7. Over-reliance — user trusts incorrect AI answer
8. Privacy violations — sensitive info in answers
9. Service degradation — LLM provider outage
10. Cost explosion — uncontrolled usage growth
11. Security — adversarial prompts extracting confidential
12. Compliance — Art. 22 violations, GDPR violations
13. Worker displacement — replacing knowledgeable staff
14. Misuse — usage outside policy
15. Vendor lock-in — dependency on OpenAI

Trampas comunes

Trampa 1 — Listar genérico, no específico. "Risk: bias" → genérico. "Risk: gender disparity in tone of responses to support queries" → específico.

Trampa 2 — Olvidar indirect stakeholders. Solo pensás en users que click el botón. Pero customers of users, society, etc. también pueden ser affected.

Trampa 3 — Risks identificados pero never mitigated. Map identifica, Manage (próximo paso) mitiga. No skip ahead sin first identifying.

Trampa 4 — Document que se hace una vez. Map debe actualizarse cuando el sistema cambia significativamente.

Ejercicio

Para tu sistema (Knowledge Assistant del Capstone):

  1. Drafteá las 5 categorías de Map
  2. Lista mínimo 10 risks identificados
  3. Categorízalos en las 7 risk categories (technical, operational, privacy, etc.)
  4. Priorizalos: critical, high, medium, low
Ver solución (priority overview)

Top priority risks:

  • Critical: tenant data leakage, hallucination affecting decisions
  • High: bias in responses, Art. 22 violations, training data leakage
  • Medium: cost explosion, vendor lock-in, stale KB
  • Low: worker displacement (consider but distant)

Priority criteria:

  • Severity × Likelihood × Visibility (audit/regulator)
  • Critical → must mitigate before deployment
  • High → mitigate in first sprint
  • Medium → plan for next quarter
  • Low → monitor

Resumen

Aprendiste:

  • ✅ 5 categorías de Map (Context, Categorization, Capabilities/Risks, Impacts, Risk identification)
  • ✅ 7 risk categories específicas para AI
  • ✅ Aplicación concreta a Knowledge Assistant
  • ✅ Trampas: generic, indirect stakeholders, sin updates

Checkpoint: si tenés un Map document con risks identified y prioritized, estás listo para Measure.

Siguiente cápsula

04 — NIST AI RMF: Measure function. Una vez mapped, medís los risks identified con metrics y testing.

Recursos

  1. NIST AI RMF Playbook — Map.
  2. AI Incident Database — for inspiration on risks.
  3. Partnership on AI — Tenets — risk thinking.
  4. Bias Audit Toolkit — your M2 deliverable.
← AnteriorNIST AI RMF: Govern FunctionSiguiente →NIST AI RMF: Measure Function